From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models

Hugging Face Daily Papers Papers

Summary

This paper proposes a staged training approach for vision-language models that separates visual perception, visual reasoning, and textual reasoning into distinct stages. The method improves visual reasoning accuracy while reducing reasoning trace length, demonstrating that stronger perception reduces the need for excessive reasoning.

Recent advances in vision-language models (VLMs) emphasize long chain-of-thought reasoning; yet, we find that their performance on visual tasks is primarily limited by a lack of visual perception as opposed to reasoning itself. In this work, we systematically study the interplay between perception and reasoning in VLM post-training by decomposing their capabilities into three separate training stages: visual perception, visual reasoning, and textual reasoning, incorporating specialized training data. We demonstrate that visual perception (a) requires targeted optimization with specialized data; (b) serves as a fundamental scaffold that should be solidified through staged training before refining visual reasoning; and (c) is more effectively learned via RL than caption-based SFT. Our experiments across multiple VLMs demonstrate that staged training consistently improves both visual perception and reasoning performance over merged training. Notably, models trained with our approach achieve 1.5% higher reasoning accuracy with 20.8% shorter reasoning traces, suggesting that superior perception reduces the need for excessive reasoning. Furthermore, we show that this capability-based staging represents a new curriculum dimension orthogonal to traditional difficulty-based curricula, and combining both yields further additive gains. Our staged-training models achieve superior performance among open-weight VLMs, establishing advanced results on several visual math and perception (e.g., +5.2% on WeMath and +3.7% on RealWorldQA) tasks compared with the base counterpart.
Original Article
View Cached Full Text

Cached at: 05/25/26, 10:37 AM

Paper page - From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models

Source: https://huggingface.co/papers/2605.20177

Abstract

Staged training approaches that separately optimize visual perception, visual reasoning, and textual reasoning in vision-language models outperform unified training methods, leading to improved performance on visual reasoning tasks.

Recent advances invision-language models(VLMs) emphasize long chain-of-thought reasoning; yet, we find that their performance on visual tasks is primarily limited by a lack ofvisual perceptionas opposed to reasoning itself. In this work, we systematically study the interplay between perception and reasoning in VLM post-training by decomposing their capabilities into three separate training stages:visual perception,visual reasoning, andtextual reasoning, incorporating specialized training data. We demonstrate thatvisual perception(a) requires targeted optimization with specialized data; (b) serves as a fundamental scaffold that should be solidified throughstaged trainingbefore refiningvisual reasoning; and (c) is more effectively learned via RL than caption-based SFT. Our experiments across multiple VLMs demonstrate thatstaged trainingconsistently improves bothvisual perceptionand reasoning performance over merged training. Notably, models trained with our approach achieve 1.5% higher reasoning accuracy with 20.8% shorter reasoning traces, suggesting that superior perception reduces the need for excessive reasoning. Furthermore, we show that this capability-based staging represents a new curriculum dimension orthogonal to traditional difficulty-based curricula, and combining both yields further additive gains. Our staged-training models achieve superior performance among open-weight VLMs, establishing advanced results on severalvisual mathand perception (e.g., +5.2% onWeMathand +3.7% onRealWorldQA) tasks compared with the base counterpart.

View arXiv pageView PDFProject pageGitHub4Add to collection

Get this paper in your agent:

hf papers read 2605\.20177

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper4

#### UCSC-VLAA/VLM-CapCurriculum-InternVL3-8B-Staged Image-Text-to-Text• 8B• Updated5 days ago • 49 #### UCSC-VLAA/VLM-CapCurriculum-InternVL3.5-8B-Staged Image-Text-to-Text• 9B• Updated5 days ago • 42 #### UCSC-VLAA/VLM-CapCurriculum-Qwen3-VL-8B-Staged Image-Text-to-Text• 9B• Updated5 days ago • 49 #### UCSC-VLAA/VLM-CapCurriculum-Qwen2.5-VL-7B-Staged Image-Text-to-Text• 8B• Updated5 days ago • 48

Datasets citing this paper3

#### UCSC-VLAA/VLM-CapCurriculum-Perception-Data Viewer• Updated5 days ago • 3.36k • 88 #### UCSC-VLAA/VLM-CapCurriculum-TextReasoning-Data Viewer• Updated5 days ago • 13.2k • 65 #### UCSC-VLAA/VLM-CapCurriculum-VisualReasoning-Data

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.20177 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Models

Papers with Code Trending

This paper uncovers that prolonged reasoning in vision-language models can impair perceptual grounding, causing recognition failures on basic visual questions. It proposes Vision-Anchored Policy Optimization (VAPO) to steer reasoning toward visually grounded trajectories, achieving state-of-the-art performance with the VAPO-Thinker-7B model.

Revisiting Complete Reasoning Traces for Post-Training

Hugging Face Daily Papers

This paper finds that large language models can gain reasoning improvements from truncated reasoning trajectories rather than full ones during post-training, reducing redundancy while benefiting methods like supervised fine-tuning and reinforcement learning.